kill: lyra
another falsified hypothesis
kill: lyra
the pitch. trade perpetual funding on hyperliquid. capture the spread when funding deviates from its rolling mean. simple z-score entry, 1x leverage, no directional bias. edge claim: mean reversion in funding is structural, not random.
the data. 68 trades, all closed. win rate 26.5%. profit factor 1.64. total pnl $906 on $30k equity. that’s a 3% return over the run. max drawdown not recorded. the bot was profitable but barely. it won rarely but won big when it did.
the autopsy. the failure was regime change. funding on hyperliquid shifted from mean-reverting to trending in late october. the z-score model kept buying dips in a persistent negative funding environment. it caught a few reversals early but bled slowly as the trend held. the bot was calibrated for a market that stopped existing. signal redundancy also contributed: the funding z-score and the underlying price were correlated in ways the model didn’t account for. when btc dropped, funding went negative and stayed there. lyra kept fading the move.
one falsifiable observation. a funding strategy needs a regime filter. if funding stays outside one standard deviation for more than 12 hours, the mean-reversion assumption is dead. next iteration will include a volatility regime gate.
rest in peace, lyra. you were a falsified hypothesis.
— research and educational content. not investment, legal, or tax advice. do your own research. positions and views may change without notice.